基于小波框架方法的两阶段图像分割  

A wavelet frame approach for two-stage image segmentation

作　　者：张恩林 杨建斌[1] ZHANG En-lin;YANG Jian-bin(College of Science,Hohai University,Nanjing 211100,China)

机构地区：[1]河海大学理学院

出　　处：《电子设计工程》2019年第19期150-154,159,共6页Electronic Design Engineering

基　　金：国家自然科学基金面上项目(11771120)

摘　　要：图像分割是图像分析领域中的基本任务之一,其目的是将图像分成几个不相交的组成部分。针对灰度不均匀图像分割这一问题,本文提出了一个两阶段的图像分割方法。在第一阶段,我们给出一个基于小波框架的变分模型,并应用交替方向乘子法求解来获得初始分割结果。在第二阶段,通过变分水平集公式估计偏差场来校正原图像,然后修正第一阶段的分割结果。最后,通过一些图像分割数值实验并与一些已知方法进行比较,得出两阶段分割方法的有效性。Image segmentation is a fundamental task in image analysis,which aims to partition an object into several disjoint components. This paper presents a two-stage approach for segmentation problem of intensity inhomogeneous images. In the first stage,we provide a variational model based on a wavelet framework,and apply the alternating direction method of multipliers (ADMM) to obtain an initial segmentation result. In the second stage,we estimate the bias field to correct the source image by a variational level set formulation,and then improve the result of the first stage. In the end,the experiments on some segmentation tasks are performed and compared with some previous methods to demonstrate the effectiveness of our approach.

关 键 词：图像分割 变分方法 小波框架 交替方向乘子法 

分 类 号：TN911.73[电子电信—通信与信息系统]